Most selection indices require specification of relative trait weights. Breeders often have difficulty in assigning weights to traits, although they can subjectively select genotypes based on performance for multiple traits. Retrospective selection indices describe selection already practiced in a population and quantify the relative trait weights used intuitively by a breeder. In this study, a team of breeders subjectively selected lines in 7 (Set A) and 10 (Set B) maize (Zea mays L.) populations evaluated at two sets of locations. Retrospective index weights were calculated as b = C-1s, where b is a vector of retrospective trait weights, C is the phenotypic variance-covariance matrix among traits, and s is a vector of selection differentials for the different traits. In Set A, the average retrospective index was I(A) = yield - 0.028(moisture) - 0.059(stalk lodging) - 0.036(root lodging). Based on standardized (unitless) relative weights, yield was the most important trait, followed by stalk lodging, moisture, and root lodging. In Set B, the average retrospective index was I(B) = yield - 0.009(moisture) - 0.017(stalk lodging) - 0.002(root lodging) - 0.016(smut infection) - 0.097(barrenness). The order of trait importance was yield, moisture, barrenness, stalk lodging, root lodging, and smut infection. When applied to independent data sets, 78% of the lines selected by the breeders also had the highest index values. Thus, retrospective selection indices may be useful to a breeder in multiple-trait selection. Most of the poorer genotypes in a population may be eliminated based on index values, giving the breeder more time to examine the more promising genotypes.
